During July 2025, Aparmp contributed to the CodeLinaro/onnxruntime repository by developing NPU-optimized implementations of the Conv and ConvTranspose operations with the auto_pad parameter set to VALID. This work focused on backend development in C++, integrating these operations with the QNN Execution Provider path to improve throughput and efficiency on CodeLinaro hardware. By eliminating CPU fallback for these specific operations, Aparmp reduced latency and enhanced resource utilization for Conv workloads. The changes were thoroughly documented and prepared for broader release, demonstrating a solid application of C++ and unit testing skills to address performance bottlenecks in ONNXRuntime’s NPU execution.